Brian Gibson

Institution: 
Santa Barbara City College
Year: 
2009

Scalable Parallel Computing for Photonic Research Simulations

Extensive simulation is needed to design custom photonic integrated circuit (PIC) components with wide fabrication tolerances, however the beam-propagation-method (BPM) and finite-difference-time-domain (FDTD) solutions can be time consuming and computer intensive. In order to accelerate this process we have developed and implemented software to make use of a preexisting network of computers to run discreet parallel processes.

Using C language, this software interfaces with Rsoft BPM/FDTD photonic simulation software to send commands from the master computer to the host computers via PSExec remote execution software. After the host computers have run their independent simulations, our software retrieves the data and collates it into graphs that are identical to the original simulation software.

The result is a nearly linear increase in speed with every computer that is added to the process in spite of some added "overhead" processes that are required for each independent simulation. While using a benchmark simulation, one computer required sixty four minutes to complete the process. Two computers required thirty three minutes to complete the same simulation.

UC Santa Barbara Center for Science and Engineering Partnerships UCSB California NanoSystems Institute